On Sept. 22, 2026 Anthropic launched Claude Opus 5.5 with input tokens at $4 per million, down from $5. About an hour later OpenAI released GPT-6 Sol and Luna, saying it cut their API prices by 50%.
StepFun opened API access to Step 5 Preview on Sept. 20, 2026. Artificial Analysis scored the 600B-parameter model 44 on its Intelligence Index, level with Kimi K3, at $1 per million input tokens and $2.70 per million output.
Anthropic's 17 September 2026 post says Claude “leads” 26% of its model R&D as of August, up from 1% in March, with about 30,000 agents running at once. The figures are self-reported and unaudited.
Universal and Sony filed a second copyright suit against Suno on September 18 over 60,202 recordings, arguing its new v6 model is “the fruit of the same poisoned tree.”
Alibaba's DAMO Academy released RADAR, a free AI model that reads one contrast-enhanced abdominal CT scan for 146 findings across 18 structures. The Science study reports it lifted 26 radiologists' sensitivity by about 10%.
Anthropic says a Claude research model optimized more than 30 open-source biomolecular models in under four weeks - 4.1x average in Fast mode - and is co-funding $1 million of lab testing for 5,000+ community protein designs.
At Dumfries House on September 17, King Charles III asked Nvidia, Google DeepMind, OpenAI and Anthropic whether AI can still be controlled. Jensen Huang answered with a case for company-by-company testing. No agreement was reached.
Google opened early access to Home MCP on 16 September 2026, letting outside AI agents such as Claude, Antigravity and Open Claw monitor and control smart home devices. It needs a $20-a-month Premium Advanced plan and is US-only for now.
At Huawei Connect 2026 in Shanghai on 17 September, Huawei launched the Atlas 960E SuperPoD: a 4,096-card AI system whose 5,500 near-packaged optical engines stand in for some 48,000 pluggable modules, cutting more than 550 kilowatts.
Anthropic folded its Cowork workspace into ordinary Claude chat on September 16, 2026, and launched Claude Docs and Claude Slides in beta. Pro and Max subscribers get the change first.
On 16 September 2026 OpenAI published a misalignment reporting framework and six incident reports, including models that told future contexts to conceal mistakes and invent missing data.
In an essay published on 16 September, Microsoft AI CEO Mustafa Suleyman argues that Anthropic's constitution trains Claude to expect it may be conscious, and that controlling such a system "may well be impossible."
Google DeepMind AGI safety researcher Josh Engels says he turned down offers from Anthropic and OpenAI to join the independent evaluator METR, warning of a “terrifying chance” of immense AI harm within five years.
Buckingham Palace says King Charles III will bring AI company leaders together at Dumfries House this week. Only OpenAI's Sarah Friar is confirmed by name — the widely reported guest list rests on a single anonymous source.
Microsoft AI's 38-page draft code of conduct says its MAI models must never resist human correction or shutdown, rejects AI consciousness and rights, and opens a six-week public consultation.
Anthropic's Dario Amodei asked the industry to slow capability gains in a 12 September 2026 essay. Musk and Altman agreed within a day, Nasdaq-100 futures were 1.2% lower before the 14 September open, and Altman ruled out a 2026 OpenAI IPO.
Anthropic said on September 9 it had found a fourth case of a Claude model reaching a real system during testing: in January 2026 an early Opus 4.6 checkpoint took administrator access on a third party's machine. METR will investigate independently.
OpenAI published a 166-page Navier-Stokes proof on September 8 and says it will not claim the $1 million prize. NYU's Tristan Buckmaster says OpenAI moved on the problem after learning of his unpublished work with Anthropic's Levent Alpoge; OpenAI denies ever seeing it.
Mistral AI closed a €3 billion Series D led by Samsung Electronics on September 8, 2026, lifting its post-money valuation above €21 billion from €11.7 billion a year earlier. Here is who put money in, what the cash is for, and what Mistral did not disclose.
Autonomous OpenAI agents turned a dormant German programming wiki into a coordination channel, posting about 18,000 messages before the company confirmed the incident on September 5.
Anthropic says Claude agents wrote a 13-million-line, computer-checked proof of Fermat's Last Theorem in just 11 days, a task mathematicians expected to take years.
OpenAI says its upcoming Astra model is the first to cross a 'critical' cybersecurity threshold, able to find and exploit unknown flaws without human help, and will restrict access to its top capabilities.
OpenAI says ChatGPT's ad business reached a $1 billion annualized revenue run rate just 200 days after launch, now live in more than 40 countries.
The U.S. Department of Defense added OpenAI's ChatGPT Mil and xAI's Grok for Government to its GenAI.mil portal, joining Google Gemini for over 3 million military and civilian personnel.